As is known in the art, multiuser detection algorithms are increasingly being considered for wireless communication systems to allow multiple transmitters to “collide” in the RF medium as viewed by the receivers for which these transmissions are targeted. In other words, multiuser detection receivers allow transmitters to be allocated the same frequency band at the same time and/or to allow collisions to occur in a packet-based multiple access wireless communication protocol.
In many cases, the addition of the multiuser detector in the receiver, along with a medium access control scheme or other protocol rules will enable such a co-channel interference tolerant radio to work. However, the quality of service (QoS) is difficult to maintain in this new paradigm of interference multiple access (IMA) when the distances between transmitters and receivers and/or the channel quality is dynamically changing. The underlying issues associated with these difficulties are a mix of both familiar problems (similar to those we understand well from typical, main stream, collision avoidance communication systems) as well as those that are unique only to IMA.
One unique problem of IMA receivers occurs when the dynamic range between the interfering signals is very large. In theory, MUD receivers should work extremely well in such a situation because even the simplest suboptimal MUD algorithm can take advantage of the inefficient high power signal. Specifically, in many situations, the high power interfering signal's rate is very low compared to its capacity potential rate based upon its very high, received power. In such a case, this high power signal can, in theory, be detected, recreated, and stripped off to reveal the low power signal underneath.
In an actual system, such as the generic state of the art MUD receiver shown in FIG. 1, the received signal, before it is fed to the MUD processing algorithm, must be collected by the receiver front end. In the receiver front end, comprised of lines and units labeled 340-750 in the figure, the signal power must be adjusted so as not to saturate the receiver. This is done by automatic gain control (AGC) unit 600. After the signal's power is reduced by the AGC, it is sampled by an analog-to-digital converter (A/D) unit 700. The A/D converter unit 700 turns the analog signal into a series of numbers or samples output on line 750. Each of these samples is represented as a block of bits (ones and zeros), where the number of bits in each block used to represent the value of a single sample has been pre-determined and is limited, typically to 8 bits per sample (bps), 12 bps, or 16 bps.
In addition to the AGC and ND operations, the processing that is done within the parameter estimation unit 800 and the MUD processing unit 900 is done with a limited precision on the number of bits per value. This is known in the art of hardware implementations of processing algorithms as fixed precision arithmetic. It is the combination of the AGC, A/D, and fixed precision arithmetic that causes a significant noise-like effect to be applied to the lower power signal.
FIG. 2 is a diagram illustrating a network scenario in which two nodes (node a and node b) endeavor to communicate with one another using a channel that is already in use by two other nodes (node A and node B). Nodes A and B may be communicating with one another at a relatively high power level and nodes a and b may desire to communicate at a much lower power level. As shown, at a particular point in time, node b may receive two signals, a higher power signal received from node A and a lower power signal received from node a. These signals are effectively added together at the receive antenna of node b. The signal from node A represents an interference signal at node b (i.e., an unwanted signal) and the signal from node a is the desired signal. This is one example of a situation where a nearby transmitter might overpower other local communications. Another example might occur in a system where power control is not possible, and where distances could be small or great between transmitting and receiving nodes, and could change over time depending upon parameters such as changing location of transmitters and/or receivers, as well as changes in the destination node(s) for each packet transmission.
If the two signals received at node b are many orders of magnitude different in received power (e.g., greater than 30 dB apart), then the combined effect of the AGC and the A/D can result in the lower of the two signals being “buried” in the combination of receiver noise and a noise-like result that happens when coarse quantization is applied during the sampling process.
Specifically, the received signal at node b may be modeled as coming from four different sources: r=s+n1+n2+n3, where s is the signal of interest and n1 is energy from other signals in the environment. Both s and n1 are present at the antenna. The other two noise terms are n2, the receiver analog noise, and n3, the receiver A/D digitization noise. The environment noise, n1, is pushed down by the AGC along with the received signal, s. The other two terms, n2 and n3, are not pushed down by the AGC. Since the AGC is upfront in the receiver chain, n2 includes impacts from the tuner and filters of the receiver. If the receiver has several stages of downconversion (e.g., one or more IF downconversion stages), there will be multiple iterations of tuners/filters, all of which contribute to n2. This means that even if the lower power signal (signal a) were received at the antenna with enough power to be successfully demodulated in the absence of the higher power signal (signal A), once it is combined with the higher power signal, the AGC will push down the total received signal power to a point where the power level of the lower power signal is dramatically lower than its received level. Even if the higher power signal could be stripped off perfectly from this post AGC signal, the rate of the information attempting to be conveyed by the lower power signal will likely be too high compared to the capacity rate dictated by the actual (post-AGC) received signal to noise ratio (SNR) of this signal.
It is possible that the effect of the AGC is not so severe as to cause the received SNR of the lower signal to be too low relative to its rate. In this case, the noise term that will have the most effect is n3, the noise that models the effect of signal quantization. Specifically, after the lower power signal has been reduced in received power by the AGC process, there may simply not be enough bits per sample to represent the small variations that are due only to the lower power of the two interfering signals. Furthermore, the processing that is performed as part of the MUD algorithm must be implemented with limited precision arithmetic. The limited precision adds yet another effect that would not cause a problem if the signals were closer in received power, but that will cause a very noticeable and detrimental effect on the lower power signal when their dynamic range (the difference between the two signal's received powers) is many orders of magnitude.
There are two conventional alternatives to using a MUD receiver to solve the interference problem. The first is to use error correction coding, treating the interference as noise. Even without the effect of the AGC-plus-A/D described above, for cases of high power interference, treating the interference as unstructured noise will result in a very low SNR for the signal of interest, which leads to a extremely low rate link. When we add the effect of the AGC-plus-ND which pushes the lower power signal of interest even lower, down into the noise, link closure, even at very low rates, is no longer possible.
The second alternative in this case is to use the spatial diversity that typically exists among interfering signals. It is very common in commercial wireless systems for the interfering transmitters to be located at different angles of arrival as seen by the receiver. If the receiver is equipped with multiple antennas, as shown in FIG. 3, the receiver could employ any number of digital beamforming algorithms to direct a strong beam toward one of the transmitters and to direct very deep nulls toward the other transmitters.
This digital beamform-only solution requires high-cost technology, and sometimes high computational complexity for needed processing algorithms, hardware devoted to the beamforming processing, as well as multiple antennas and the corresponding receiver ports. In addition, to attempt to provide sufficient quality of service under many typical conditions, the number of antenna elements needs to be greater than the total number of anticipated interfering transmitters. This is unrealistic for most size/weight/power requirements of most wireless communication receivers. Even with all these necessary components, there are still many cases in which the signals are not spatially separable to the level required for reliable communication. Moreover, the limited dynamic range issue is still present in conventional state of the art digital beamforming approaches, such as illustrated in FIG. 3.
A much lighter and cheaper, but less effective, solution could be built with as few as two receive antennas and by using analog beamforming instead of digital beamforming. Such a receiver would provide some capability in adjusting the received powers of interfering signals, but alone, would almost always suffer unreliable links due to the unsuppressed co-channel interference.